ShotSpotter Sample Live Fire Test Data

This dataset comprises gunshot audio and supporting data released as part of ShotSpotter Tech Note 098, "Precision and accuracy of acoustic gunshot location in an urban environment".

The data derive from a series of live fire tests of the ShotSpotter Respond gunshot location system conducted in Pittsburgh, PA on December 18th, 2018 by the Pittsburgh Bureau of Police. ShotSpotter uses live fire tests to validate that the deployed sensor density is appropriate for the community in question, and to ensure the system is ready for production use.

Source data from the Pittsburgh tests is comparable to what is submitted as evidence in a criminal trial. These materials comprise time-stamped acoustic recordings, time- and location-stamped output of the pulse processing algorithms, and tagged pulse sets suitable for acoustic multilateration. These data permit other investigators to review or extend the current work.

This dataset is released under the Creative Commons CC BY 4.0 license.

For more about ShotSpotter, see https://www.shotspotter.com.

Data Released

The dataset comprises audio files recorded by ShotSpotter's acoustic sensors, metadata that describe audio files, and metadata associated with the test itself such as the surveyed location of the firing position. Audio files from up to six reporting sensors are provided for each test.

Both the sensor positions and the firing position have been anonymized by converting location of each sensor from latitude-longitude-elevation to an arbitrary Cartesian x-y-z coordinate system. Each firing position has a different arbitrary origin.

Primary Data

Data provided on each test is as follows:

Sensor Audio: test_{xxx}.audio.{sensor serial number}.wav

These files are 12 kHz 16-bit mono uncompressed audio files in WAV format. The timestamp of the first sample of the file is given in the associated metadata.

Sensor Metadata: test_{xxx}.audio.{sensor serial number}.json

Metadata associated with audio files, as follows:

field	description
serial_number	sensor serial number
timestamp	UTC timestamp of the start of the first sample
location	xyz sensor location in arbitrary coordinate system (meters)
description	human-readable description of the audio
original_file_md5	checksum of the original (non-anonymized) file in ShotSpotter's private data repository.

Survey Location: test_{xxx}.survey.json

Surveyed location of the firing test, as determined by a GPS-enabled smartphone. Accuracy of the survey location is approximately 2.5 meters.

field	description
test_id	test id, sequential from start of test
firing_position	each firing position has a different arbitrary origin
num_rounds	number of shots fired in the test
firearm	type of firearm (one of three)
approximate_time_utc	UTC timestamp recorded by the program manager. Not an accurate discharge time
survey_location	xyz discharge location in arbitrary coordinate system (meters)
description	human-readable description of test

Weather: test_{xxx}.weather.json

Weather data, reported from the nearest available NWS METAR station.

field	description
fluid	always "air"
temperature	in deg C
speed	windspeed in m/s
direction	wind direction, as a from azimuth
observation_time	as reported by METAR station
source	reporting station

Derived Data

The data above comprise the primary dataset; it is possible to analyze the provided audio files, identify the gunshot pulses, and compare the computed locaton with the survey location using only the data above.

The dataset also includes two additional files describe the output of ShotSpotter's signal processing and pulse grouping algorithms. These may be useful to researchers who are primarily interested in multilateration accuuracy.

Pulse Data: test_{xxx}.allpulses.json

Pulse data reported by ShotSpotter sensors. This file includes pulses from sensors for which audio data is not provided.

field	description
pulse_uuid	uuid
serial_number	serial number of the reporting sensor. There are not audio files associated with all pulses in this file.
arrival_time_utc	pulse arrival time, zero-crossing point at start of pulse
location	xyz sensor location in arbitrary coordinate system (meters)
location_error	sensor's self-described position error (meters)
signal_to_noise_dB	ratio of peak value of gunshot pulse to pre-pulse noise
power_dB_SPL	peak power of the pulse, in dB SPL

Pulse Selection Data: test_{xxx}.nlos-tuples.json

Sets of pulses from test_{xxx}.allpulses.json associated with each individual shots, as selected by ShotSpotter algorithms.

field	description
shot_index	index within test, (0..2)
pulse_uuids	foreign keys to test_{xxx}.allpulses.json

Tests Conducted

There were a total of nine firing positions. Eighteen tests were conducted at each firing position, for a total of 162 tests.

Tests were conducted in the following sequence at each firing position:

.45 caliber automatic handgun

---

• 3 rounds
• 3 rounds
• 3 rounds
• 1 round
• 1 round
• 1 round

.40 caliber automatic handgun

---

• 3 rounds
• 3 rounds
• 3 rounds
• 1 round
• 1 round
• 1 round

9 mm automatic handgun

---

• 3 rounds
• 3 rounds
• 3 rounds
• 1 round
• 1 round
• 1 round

Misc

Comments or questions can be directed to rcalhoun@shotspotter.com.

Last update 29 July 2021.